Managing Hardware Non-Determinism in the Age of AI and Regulations.. as Usual

Systems Engineering is often misunderstood as the pursuit of absolute determinism - however, pure determinism is a myth. Unpredictability and variance are inherent in our systems, from fundamental physics uncertainty to variety in engineering work, production tolerances or user behaviour, managing non-determinism essential in engineering.

Standards like ISO 13485 in MedTech or ECSS in Space do not enforce zero variance; they provide frameworks and processes like FMECA and risk management (ISO 14971) to bound it.

AI introduces a novel semantic layer to this spectrum. We will explore three core theses:

• AI as an SE Co-Pilot: Utilizing LLMs to conduct audits, refine requirements, and explore variance through automated hazard scenario generation.
• Subsystem Bounding and internal Interfaces: Non-deterministic AI modules can be seen like human operators. By applying appropriate checks, we ensure operations stay within statistical confidence levels.
• The AI-Agent User: Designing hardware for an era where system "users" are AI agents, exhibiting their own spectrum of (non)determinism.

This presentation looks to organize our thinking for the new paradigm of integration of AI into critical and regulated environments and how to use the existing Systems Engineering concepts to confidently leverage AI without compromising performance, integrity or compliance.